Semi-supervised Spectral Clustering Combined with Bayesian Decision

This paper proposes a semi-supervised spectral clustering algorithm combined with Bayes decision, for Low stability and accuracy of spectral clustering algorithm. This method is clustered according to color, texture and spatial characteristics of the image. It first adjusts the similarity matrix by distance learning methods based on Bayes decision to improve clustering distribution of feature vectors, Then, we use the constrained K-means algorithm to cluster adjusted feature vectors to further improve the stability and accuracy of results. The synthetic texture image and natural image segmentation experiments show that this method has significantly improved stability and accuracy than traditional spectral clustering.